All product areas

07 / Blackbox

The explanation behind every finished change.

Blackbox keeps the sequence that produced a result: planning, dispatch, tool activity, verification, recovery, decisions, and the final synthesis.

A real Sourceweave workspace with its Blackbox event history visible Actual Blackbox workspace capture

WHAT IT CHANGES

Less reconstruction.
More visible progress.

01

A durable run timeline

See when assignments started, blocked, retried, completed, and entered review.

02

Evidence bound to decisions

Keep verification output and user approvals beside the state change they allowed.

03

Recovery without guesswork

Compare attempts and resume from a known branch or checkpoint when a run needs intervention.

04

Filter to the event that matters

Narrow history by agents, commands, files, Git, verification, workspace activity, or attention-required events.

05

Compare competing attempts

Place models, effort, duration, files, verification, and timelines side by side before choosing a result.

06

Export the record

Create Markdown or JSON reports for a run or comparison without losing the structured evidence behind it.

WHAT IS BUILT IN

The working surfaces behind Blackbox.

TIMELINE REPLAY01

Play back what Sourceweave recorded

Scrub, play, pause, and step through durable events by workspace, task, and attempt without pretending to rerun the agent deterministically.

EVENT INSPECTOR02

Open the payload behind the marker

Inspect phase, source, sequence, changed files, check output, branch data, task prompts, and links back to the original agent attempt or Foundry evidence.

CHECKPOINTS03

Recover without rewriting the present

Preview a manual or automatic pre-apply checkpoint against current files, then create a recovery branch while leaving the active branch untouched.

COMPARISONS04

Choose with context

Compare two attempts, scope shared and unique files, open their changes or previews, and export a clean decision report.

HOW IT WORKS

A deliberate path through Blackbox.

  1. 01

    Capture the control loop

    Planning, worker dispatch, integrations, and review actions become structured run events.

  2. 02

    Keep the useful output

    Final agent messages remain prominent while low-level command output stays available on demand.

  3. 03

    Compare or recover

    Place attempts side by side or create a recovery branch from a known checkpoint without disturbing current work.

  4. 04

    Return with context

    Inspect a past result or failure without reconstructing the entire story from terminal history.

KEEP EXPLORING

01Built-in IDE02Workspaces03Agents04Agent Tasks05Agent Swarms06Foundry + Review08Voice